You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@tinkerpop.apache.org by "Jayanta Mondal (JIRA)" <ji...@apache.org> on 2019/03/09 01:13:00 UTC

[jira] [Updated] (TINKERPOP-2177) Streaming response immediately after authentication stops after first partial response

     [ https://issues.apache.org/jira/browse/TINKERPOP-2177?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jayanta Mondal updated TINKERPOP-2177:
--------------------------------------
    Description: 
+*Repro step*+: 

Use gremlinpython 3.4.0 to issue 'g.V()' (or any gremlin query) on a sufficiently large graph, so that the response is large enough that the server needs to send multiple partial responses (status code: 206). 

The 'g.v()' needs to be the first query immediately after establishing the connection. 

+*Behavior:*+ 

The query only returns the first partial response from the gremlin server to the user.  Even though the gremlinpython client ends up reading all the data from the gremlin server. 

 

*+Why Critical:+* 
 # This is a correctness issue from the end-user point of view.
 # The client is only getting partial data while paying the cost of running the entire query on the gremlin server.

 

+*Diagnosis:*+ 

There was a recent change in gremlinpython to stop using a recursive call to read from a steaming response. The change made the caller (def _receive(self): in connection.py) use a while loop to read the streaming response.

 

While this change is fine, on the first request to a WebSocket connection, after the authentication is done, we are still making a recursive call to read data from the response stream. Ideally, after the authentication is done, we should return control to the def _receive(self) method so that it can do the read. 

Such mixing of recursive call and while loop is causing the following behavior: 

1. If the first response immediately after the authentication is a streaming response, we are only reading the first chunk of the result. 

 

 

[https://github.com/apache/tinkerpop/commit/4c8717dd4e94ec248a959911d8c11f3b45b2d7b3#diff-c3daf7e789724ff22bcba3d02f3dc35a]

+*Fix:*+

 Add a '{color:#205081}return{color}' before the recursive call to read data after auth (see attachment below)

       --> This means that the caller now can use the while loop read the streaming response.

       --> Otherwise the caller gets back a status code 'None' and thinks that it is not a partial response. 

 

 

 

  was:
+*Repro step*+: 

Use gremlinpython 3.4.0 to issue 'g.V()' (or any gremlin query) on a sufficiently large graph, so that the response is large enough that the server needs to send multiple partial responses (status code: 206). 

The 'g.v()' needs to be the first query immediately after establishing the connection. 

+*Behavior:*+ 

The query only returns the first partial response from the gremlin server to the user.  Even though the gremlinpython client ends up reading all the data from the gremlin server. 

 

*+Why Critical:+* 
 # This is a correctness issue from the end-user point of view.
 # The client is only getting partial data while paying the cost of running the entire query on the gremlin server.

 

+*Diagnosis:*+ 


There was a recent change in gremlinpython to stop using a recursive call to read from a steaming response. The change made the caller (def _receive(self): in connection.py) use a while loop to read the streaming response.

 

While this change is fine, on the first request to a WebSocket connection, after the authentication is done, we are still making a recursive call to read data from the response stream. Ideally, after the authentication is done, we should return control to the def _receive(self) method so that it can do the read. 

Such mixing of recursive call and while loop is causing the following behavior: 

1. If the first response immediately after the authentication is a streaming response, we are only reading the first chunk of the result. 

 

 

[https://github.com/apache/tinkerpop/commit/4c8717dd4e94ec248a959911d8c11f3b45b2d7b3#diff-c3daf7e789724ff22bcba3d02f3dc35a]

+*Fix:*+

 Add a '{color:#205081}return{color}' before the recursive call to read data after auth.

       --> This means that the caller now can use the while loop read the streaming response.

       --> Otherwise the caller gets back a status code 'None' and thinks that it is not a partial response. 

 

 

 


> Streaming response immediately after authentication stops after first partial response
> --------------------------------------------------------------------------------------
>
>                 Key: TINKERPOP-2177
>                 URL: https://issues.apache.org/jira/browse/TINKERPOP-2177
>             Project: TinkerPop
>          Issue Type: Bug
>          Components: python
>    Affects Versions: 3.4.0
>         Environment: Independent of the environment. We could repro the issue both on Windows and Linux. 
>            Reporter: Jayanta Mondal
>            Priority: Critical
>         Attachments: gremlinpythonbug.png
>
>   Original Estimate: 3h
>  Remaining Estimate: 3h
>
> +*Repro step*+: 
> Use gremlinpython 3.4.0 to issue 'g.V()' (or any gremlin query) on a sufficiently large graph, so that the response is large enough that the server needs to send multiple partial responses (status code: 206). 
> The 'g.v()' needs to be the first query immediately after establishing the connection. 
> +*Behavior:*+ 
> The query only returns the first partial response from the gremlin server to the user.  Even though the gremlinpython client ends up reading all the data from the gremlin server. 
>  
> *+Why Critical:+* 
>  # This is a correctness issue from the end-user point of view.
>  # The client is only getting partial data while paying the cost of running the entire query on the gremlin server.
>  
> +*Diagnosis:*+ 
> There was a recent change in gremlinpython to stop using a recursive call to read from a steaming response. The change made the caller (def _receive(self): in connection.py) use a while loop to read the streaming response.
>  
> While this change is fine, on the first request to a WebSocket connection, after the authentication is done, we are still making a recursive call to read data from the response stream. Ideally, after the authentication is done, we should return control to the def _receive(self) method so that it can do the read. 
> Such mixing of recursive call and while loop is causing the following behavior: 
> 1. If the first response immediately after the authentication is a streaming response, we are only reading the first chunk of the result. 
>  
>  
> [https://github.com/apache/tinkerpop/commit/4c8717dd4e94ec248a959911d8c11f3b45b2d7b3#diff-c3daf7e789724ff22bcba3d02f3dc35a]
> +*Fix:*+
>  Add a '{color:#205081}return{color}' before the recursive call to read data after auth (see attachment below)
>        --> This means that the caller now can use the while loop read the streaming response.
>        --> Otherwise the caller gets back a status code 'None' and thinks that it is not a partial response. 
>  
>  
>  



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)